Transcript Document

Panel on Hedonic Price
Indexes
Robert J. Gordon
Northwestern University and NBER
Vancouver, CRIW Conference on Price Index
Concepts and Measurement, June 28, 2004
7/18/2015
1
Difference in Emphasis – Longrun History vs. Short-run Index
Number Construction
• In historical research
– Data doesn’t have to be contemporaneous, it doesn’t
matter how long it takes to arrive
– We keep revising history when new data arrive
– Emphasis on yearly or even decade-by-decade
changes, not monthly
• Over long time periods, cumulative changes in
both quality and price can be large
7/18/2015
2
How can we check price indexes
for plausibility?
• Over decades, price indexes and implied quality indexes
can drift away from plausibility
• But for some products (not all), can compare “closely
similar models” across decades
• Catalog photos and specifications, Consumer Reports
– Doesn’t help with computers
• Vancouver evokes thoughts
– Wimpy Vancouver Chevy taxi vs. the 1949 Buick
Roadmaster. CPI implies 3x quality
– Vancouver at sunset, classic 1955 Chevrolet convertible
7/18/2015
3
Why Doesn’t Every Study Exhibit
Quality and Price Indexes
Side-by-Side?
• Every hedonic study has the raw material
for a quality index
• Quality = raw price divided by hedonic
price index
• Quality index can be checked for
plausibility (apparel study, memo for
rental shelter study)
7/18/2015
4
Another Neglected Theme
• For some products operating costs are large
relative to initial capital cost
• Fuel economy and energy efficiency can shift
but need to be valued separately (different fuel
price regimes)
– Autos
– Refrigerators
– Air conditioners
7/18/2015
5
Stark Differences between
Products
• For mainframe computers and PCs, quality
change is so rapid that small differences in
measurement methodology don’t matter
• For apparel, price changes are so big over
decades and quality change is so small that we
can get away with no quality adjustments at all
• Many products are in between, no alternative
but to do the hard work and get as much data
as possible
7/18/2015
6
The Hedonic Literature is
Dominated by Durable Goods
• Durable Goods: Continual Arrival of New Products, Quality
Improvements in Old Products
• But much of the market basket consists of nondurables,
services.
– Lawnmowing services, two inventions in the last two
centuries (better fertilizers and weed killers?)
– Barber shops, just one invention in two centuries?
• Court, Griliches on autos
– Auto issues dominated the hedonic literature in the 1960s,
1970s
7/18/2015
7
In the 1960s Zvi and Jack Triplett
wrestled with the core issues
about autos
• Early recognition of hedonic limitations
– Physical characteristics vs. performance
– We wanted speed, power, comfort, fuel economy
– What we had was weight, length, horsepower
• General issue: shifting ratio of utility-enhancing
performance attributes to physical attributes
7/18/2015
8
The “Luxury Model” Problem
• Coefficient on weight overstated
• Reflected “prestige” attribute, 10% extra weight
might add 30% to price
• Overstated coefficient on weight
• The upward biased coefficient on weight
caused quality change to be overstated, price
change
7/18/2015
9
Solution for Luxury Models?
• Make Effects
• Zvi first pointed out compact car bias in mixed
samples (1959, 1960)
• Simple solution, restrict sample to same makes
• Griliches-Ohta (1976) – extensive attention to
make-effect dummies
• This has been picked up in studies of TV and
PCs. The “Sony” effect
7/18/2015
10
Left-out Variables, Zvi had nailed
this issue intuitively
• Biggest example, very important after 1973: fuel economy
• Fuel economy negatively correlated with weight
• Couldn’t estimate its own coefficient
• Solution: Wilcox, separate regression explaining FE as a
function of weight, time, and other characteristics
– We can actually do a pretty good job of comparing the
1949 Buick Roadmaster with the wimpy Vancouver Chevy
taxis
– Value the fuel economy, value the air bags
7/18/2015
11
Reasons hedonics were
necessary but not sufficient
• Had to add auxiliary regression for fuel economy
• Weight-saving innovations, downsizing
• Triplett concluded much later: autos “too complicated” for
hedonic studies
• Difficulty of estimating plausible coefficients on
accessories, e.g., air conditioning
• No alternative to going beyond simple hedonic regressions
7/18/2015
12
Examples where hedonics work
well and badly
• Autos – too complex
• Commercial aircraft, not enough models
– Net revenue criterion
• Computers, ideal but becoming more complex
– OK, speed and memory
– Size, clarity of screen. CDs vs. floppies, ethernet vs.
phone cord (my delightful experience in Vancouver)
• Complementarities (web/PC, I-state/auto)
7/18/2015
13
The Biggest CPI Component of
all: Rental Shelter
• Good data getting better: 1975-2003 American
housing survey
• Huge sample, lots of quality characteristics
• Controls for quality and non-quality
characteristics
– Quality: sq feet, # of rooms, AC
– Non-quality: regional location, implied land rent
7/18/2015
14
The Hard Part: 1914-1975
• 1975-2003 Hedonics provides coefficients to
evaluate earlier changes in quality (sq ft, # of
rooms, # of bathrooms)
• Comparisons over decades
• How did we get from there to here?
– Start with no indoor plumbing, no central heating.
How much could that have been worth?
7/18/2015
15
Conclusions
• #1 Hedonics necessary but not sufficient
• #2 Everyone should be producing quality indexes, not just
price indexes
• #3 Every historical study should have an obligation to
check the implicit quality index for plausibility
• #4 Huge differences across products
– In ratio of price change to quality change
– In the adequacy of hedonics and beyond-hedonic
adjustments, e.g., fuel economy and energy efficiency
7/18/2015
16